VPMADD52LUQ X5, X9, K1, X24                        // 6262b509b4c5
	VPMADD52LUQ X31, X9, K1, X24                       // 6202b509b4c7
	VPMADD52LUQ X3, X9, K1, X24                        // 6262b509b4c3
	VPMADD52LUQ 15(R8), X9, K1, X24                    // 6242b509b4800f000000
	VPMADD52LUQ (BP), X9, K1, X24                      // 6262b509b44500

  }

  private enum SmallEnum {
    X0,
    X1,
    X2
  }

  private enum MediumEnum {
    X0,
    X1,
    X2,
    X3,
    X4,
    X5,
    X6,
    X7,
    X8,
    X9,
    X10,
    X11,
    X12,
    X13,
    X14,
    X15,
    X16,
    X17,
    X18,
    X19,
    X20,
    X21,
    X22,
    X23,
    X24,
    X25,
    X26,

	VMOVQ		R4, V3.B[1]     // 8384eb72
	VMOVQ		R5, V4.B[3]     // a48ceb72
	VMOVQ		R6, V5.H[2]     // c5c8eb72
	VMOVQ		R7, V6.W[2]     // e6e8eb72
	VMOVQ		R8, V7.V[1]     // 07f5eb72
	XVMOVQ		R7, X9.W[2]     // e9c8eb76
	XVMOVQ		R8, X10.V[2]    // 0ae9eb76

	// Duplicate general-purpose register to vector
	VMOVQ		R4, V2.B16      // 82009f72
	VMOVQ		R5, V3.H8       // a3049f72
	VMOVQ		R6, V4.W4       // c4089f72

	VPERMB -7(DI)(R8*2), X2, K1, X23                   // 62a26d098dbc47f9ffffff
	VPERMB X26, X9, K1, X23                            // 628235098dfa
	VPERMB X19, X9, K1, X23                            // 62a235098dfb
	VPERMB X0, X9, K1, X23                             // 62e235098df8
	VPERMB 7(SI)(DI*4), X9, K1, X23                    // 62e235098dbcbe07000000
	VPERMB -7(DI)(R8*2), X9, K1, X23                   // 62a235098dbc47f9ffffff

	VXORPS X3, X21, K2, X9                             // 6271540257cb
	VXORPS 15(DX)(BX*1), X21, K2, X9                   // 62715402578c1a0f000000
	VXORPS -7(CX)(DX*2), X21, K2, X9                   // 62715402578c51f9ffffff
	VXORPS X11, X1, K2, X9                             // 6251740a57cb
	VXORPS X31, X1, K2, X9                             // 6211740a57cf

	VRCP28SD X2, X24, K2, X9                           // 6272bd02cbca or 6272bd22cbca or 6272bd42cbca
	VRCP28SD X27, X24, K2, X9                          // 6212bd02cbcb or 6212bd22cbcb or 6212bd42cbcb
	VRCP28SD X26, X24, K2, X9                          // 6212bd02cbca or 6212bd22cbca or 6212bd42cbca

	VPSHRDVW 7(SI), X8, K7, X9                         // 6272bd0f728e07000000
	VPSHRDVW X2, X9, K7, X9                            // 6272b50f72ca
	VPSHRDVW X31, X9, K7, X9                           // 6212b50f72cf
	VPSHRDVW X11, X9, K7, X9                           // 6252b50f72cb
	VPSHRDVW (AX), X9, K7, X9                          // 6272b50f7208

	VPDPWSSD X2, X2, K3, X9                            // 62726d0b52ca
	VPDPWSSD X27, X2, K3, X9                           // 62126d0b52cb
	VPDPWSSD X26, X2, K3, X9                           // 62126d0b52ca
	VPDPWSSD (SI), X2, K3, X9                          // 62726d0b520e
	VPDPWSSD 7(SI)(DI*2), X2, K3, X9                   // 62726d0b528c7e07000000
	VPDPWSSD X2, X24, K3, X9                           // 62723d0352ca

	VPBROADCASTMW2D K7, X11                            // 62727e083adf
	VPBROADCASTMW2D K6, X11                            // 62727e083ade
	VPBROADCASTMW2D K7, X9                             // 62727e083acf
	VPBROADCASTMW2D K6, X9                             // 62727e083ace
	VPBROADCASTMW2D K4, Y22                            // 62e27e283af4
	VPBROADCASTMW2D K6, Y22                            // 62e27e283af6

	VFMADD132SD X13, X9, K7, X0                        // 62d2b50f99c5 or 62d2b52f99c5 or 62d2b54f99c5
	VFMADD132SD 17(SP)(BP*8), X9, K7, X0               // 62f2b50f9984ec11000000 or 62f2b52f9984ec11000000 or 62f2b54f9984ec11000000
	VFMADD132SD 17(SP)(BP*4), X9, K7, X0               // 62f2b50f9984ac11000000 or 62f2b52f9984ac11000000 or 62f2b54f9984ac11000000
	VFMADD132SS X9, X8, K4, X2                         // 62d23d0c99d1